Asta-Advances in Statistical Analysis最新文献

A common choice for the marginal distribution of a bivariate count time series is the bivariate Poisson distribution. In practice, however, when the count data exhibit zero inflation, overdispersion or non-stationarity features, such that a marginal bivariate Poisson distribution is not suitable. To test the discrepancy between the actual count data and the bivariate Poisson distribution, we propose a new goodness-of-fit test based on a bivariate dispersion index. The asymptotic distribution of the test statistic under the null hypothesis of a first-order bivariate integer-valued autoregressive model with marginal bivariate Poisson distribution is derived, and the finite-sample performance of the goodness-of-fit test is analyzed by simulations. A real-data example illustrate the application and usefulness of the test in practice.

In this paper, we propose a Bayesian quantile regression (QR) approach to jointly model multivariate ordinal data. Firstly, a multivariate latent variable model is used to link the multivariate ordinal data and latent continuous responses and the multivariate asymmetric Laplace (MAL) distribution is employed to construct the joint QR-based working likelihood for the considered model. Secondly, adaptive-(L_{1/2}) penalization priors of regression parameters are incorporated into the working likelihood to implement high-dimensional Bayesian joint QR inference. Markov Chain Monte Carlo (MCMC) algorithm is utilized to derive the fully conditional posterior distributions of all parameters. Thirdly, Bayesian joint relatively QR estimation approach is recommended to result in more efficient estimation results. Finally, Monte Carlo simulation studies and a real instance analysis of multirater agreement data are presented to illustrate the performance of the proposed Bayesian joint relatively QR approach.

This paper focuses on the general linear measurement error model, in which some or all predictors are measured with error, while others are measured precisely. We propose a semi-parametric estimator that works under general mechanisms of measurement error, including differential and non-differential errors. Other popular methods, such as the corrected score and conditional score methods, only work for non-differential measurement error models, but our estimator works in all scenarios. We develop our estimator by considering a family of objective functions that depend on an unspecified weight function. Using statistical error analysis and perturbation theory, we derive the optimal weight function under the small-sigma regime. The resulting estimator is statistically optimal in all senses. Even though we develop it under the small-sigma regime, we also establish its consistency and asymptotic normality under the large sample regime. Finally, we conduct a series of numerical experiments to confirm that the proposed estimator outperforms other existing methods.

Benford’s law is a particular discrete probability distribution that is often satisfied by the significant digits of a dataset. The nonconformity with Benford’s law suggests the possible presence of data manipulation. This paper introduces two novel generalized versions of Benford’s law that are less restrictive than the original Benford’s law—hence, leading to more probable conformity of a given dataset. Such generalizations are grounded on the existing mathematical relations between Benford’s law probability distribution elements. Moreover, one of them leads to a set of probability distributions that is a proper subset of that of the other one. We show that the considered versions of Benford’s law have a geometric representation on the three-dimensional Euclidean space. Through suitable optimization models, we show that all the probability distributions satisfying the more restrictive generalization exhibit at least acceptable conformity with Benford’s law, according to the most popular distance measures. We also present some examples to highlight the practical usefulness of the introduced devices.

We propose to investigate the joint dynamics of regional gross domestic product and life expectancy in Italy through Wasserstein barycenter regression derived from optimal transport theory. Wasserstein barycenter regression has the advantage of being flexible in modeling complex data distributions, given its ability to capture multimodal relationships, while maintaining the possibility of incorporating uncertainty and priors, other than yielding interpretable results. The main findings reveal that regional clusters tend to emerge, highlighting inequalities in Italian regions in economic and life expectancy terms. This suggests that targeted policy actions at a regional level fostering equitable development, especially from an economic viewpoint, might reduce regional inequality. Our results are validated by a robustness check on a human mobility dataset and by an illustrative forecasting exercise, which confirms the model’s ability to estimate and predict joint distributions and produce novel empirical evidence.

It is often of primary interest to analyze and forecast the levels of a continuous phenomenon as a categorical variable. In this paper, we propose a new spatio-temporal model to deal with this problem in a binary setting, with an interesting application related to the COVID-19 pandemic, a phenomena that depends on both spatial proximity and temporal auto-correlation. Our model is defined through a hierarchical structure for the latent variable, which corresponds to the probit-link function. The mean of the latent variable in the proposed model is designed to capture the trend and the seasonal pattern as well as the lagged effects of relevant regressors. The covariance structure of the model is defined as an additive combination of a zero-mean spatio-temporally correlated process and a white noise process. The parameters associated with the space-time process enable us to analyze the effect of proximity of two points with respect to space or time and its influence on the overall process. For estimation and prediction, we adopt a complete Bayesian framework along with suitable prior specifications and utilize the concepts of Gibbs sampling. Using the county-level data from the state of New York, we show that the proposed methodology provides superior performance than benchmark techniques. We also use our model to devise a novel mechanism for predictive clustering which can be leveraged to develop localized policies.

Considerable research has been devoted to understanding the popularity effect on the art market dynamics, meaning that artworks by popular artists tend to have high prices. The hedonic pricing model has employed artists’ reputation attributes, such as survey results, to understand the popularity effect, but the reputation attributes are constant and not properly defined at the point of artwork sales. Moreover, the artist’s ability has been measured via random effect in the hedonic model, which fails to reflect ability changes. To remedy these problems, we present a method to define the popularity measure using the artwork sales dataset without relying on the artist’s reputation attributes. Also, we propose a novel pricing model to appropriately infer the time-dependent artist’s abilities using the presented popularity measure. An inference algorithm is presented using the EM algorithm and Gibbs sampling to estimate model parameters and artist abilities. We use the Artnet dataset to investigate the size of the rich-get-richer effect and the variables affecting artwork prices in real-world art market dynamics. We further conduct inferences about artists’ abilities under the popularity effect and examine how ability changes over time for various artists with remarkable interpretations.

Decision trees constitute a simple yet powerful and interpretable machine learning tool. While tree-based methods are designed only for cross-sectional data, we propose an approach that combines decision trees with time series modeling and thereby bridges the gap between machine learning and statistics. In particular, we combine decision trees with hidden Markov models where, for any time point, an underlying (hidden) Markov chain selects the tree that generates the corresponding observation. We propose an estimation approach that is based on the expectation-maximisation algorithm and assess its feasibility in simulation experiments. In our real-data application, we use eight seasons of National Football League (NFL) data to predict play calls conditional on covariates, such as the current quarter and the score, where the model’s states can be linked to the teams’ strategies. R code that implements the proposed method is available on GitHub.

When asked to assess their opinion about attitudes or perceptions on Likert-scale, respondents often endorse the midpoint or extremes of the scale and agree or disagree regardless of the content. These responding behaviors are known in the psychometric literature as middle, extremes, aquiescence and disacquiescence response styles that generally introduce bias in the results. One of the key motivations behind our approach is to account for these attitudes and how they evolve over time. The novelty of our proposal, in the context of longitudinal ordered categorical data, is in considering simultaneously the temporal dynamics of the responses (observable ordinal variables) and unobservable answering behaviors, possibly influenced by response styles, through a Markov switching logit model with two latent components. One component accommodates serial dependence and respondent’s unobserved heterogeneity, the other component determines the responding attitude (due to response styles or not). The dependence of the responses on covariates is modelled by a stereotype logit model with parameters varying according to the two latent components. The stereotype logit model is adopted because it is a flexible extension of the proportional odds logit model that retains the advantage of using a single parameter to describe a regressor effect. In the paper, a new interpretation of the parameters of the stereotype model is given by defining the allocation sets as intervals of values of the linear predictor that identify the most probable response. Unobserved heterogeneity, serial dependence and tendency to response style are modelled through our approach on longitudinal data, collected by the Bank of Italy.

In this article, a new kind of interpretable machine learning method is presented, which can help to understand the partition of the feature space into predicted classes in a classification model using quantile shifts, and this way make the underlying statistical or machine learning model more trustworthy. Basically, real data points (or specific points of interest) are used and the changes of the prediction after slightly raising or decreasing specific features are observed. By comparing the predictions before and after the shifts, under certain conditions the observed changes in the predictions can be interpreted as neighborhoods of the classes with regard to the shifted features. Chord diagrams are used to visualize the observed changes. For illustration, this quantile shift method (QSM) is applied to an artificial example with medical labels and a real data example.